GridWork HQ
AI Pipelines

Spec Pipeline

Generate technical specifications from client requirements — with a review gate after Pass 1.

Spec Pipeline

The spec pipeline transforms client requirements into a structured specification document across three passes, with a mandatory review gate after the first pass.

Quick Reference

FieldValue
Triggerspec [client name]
DomainClient Delivery
ModelPass 1: Opus, Pass 2: Sonnet, Pass 3: Opus
Requiresclients/[name]/context.yaml and clients/[name]/brief.md
OutputThree files: spec-requirements.md, spec-design.md, spec.md
Review GateYes — stops after Pass 1 for approval before continuing

How It Works

Pass 1 — Requirements Extraction (Opus)

  • Reads all client context files
  • Extracts every functional and non-functional requirement
  • Structures using EARS notation (When/The system shall)
  • Assigns IDs (REQ-001, REQ-002, ...)
  • Groups by page/feature area
  • Output: clients/[name]/spec-requirements.md
  • STOPS HERE — you review requirements with the client before proceeding

Pass 2 — Design Document (Sonnet)

  • Reads the approved requirements
  • Designs the technical solution: page structure, component hierarchy, data model
  • Maps requirements to implementation areas
  • Output: clients/[name]/spec-design.md

Pass 3 — Task Breakdown (Opus)

  • Breaks the design into ordered, executable tasks
  • Maps tasks to specific pipeline runs
  • Estimates time per task and identifies dependencies
  • Output: clients/[name]/spec.md — the canonical spec for plan and build

The Review Gate

The spec pipeline is the only pipeline with a built-in review gate. After Pass 1, it commits the output and stops. You review the requirements with the client and explicitly trigger Pass 2 when ready. This prevents building from incorrect assumptions.

On this page